1070 overclock never getting artifacts

Wrezl25

Commendable
Nov 7, 2016
4
0
1,510
So I have been stumped for a while now. Most guides I read say bump up the core clock by 5-10MHz and when you get artifacts back it down a few notches and then keep Valley, Kumbustor or whatever program you are using to stress test running for an hour or two and if it doesn't artifact you have a stable over clock. The problem is I have never had any artifacts. The only thing that has happened is my stress tests have crashed right once I hit +253MHz. When I am at +252MHz it all runs fine. Kumbustor 3 seems to be able to hold +272MHz but crashes at +273MHz. Also changing my voltage% or power limit changed nothing as well. So my question is is this normal and my stable clock is +252MHz or is it something else I might be doing wrong?

Programs used:
MSI afterburner
Valley Benchmark
Kumbustor 3
3D Mark
GPU-z
 
Solution

NerdIT

Distinguished


Heya,

So, no matter what you do -you will eventually hit a ceiling no matter what you do. Artifacts are not an accurate way to figure out your overclock. Yes, artifacts are bad -but there is way more that goes into OC'ing GFX then artifacts..

The next to last graph/metric on GPU-Z- called "PerfCap Reason" will indicate what is happening. The most common factors will be power "Pwr" and reliability voltage "Vrel".

No matter what you do - one of these two will ultimately "cap" your potential OC. The more you OC the core, the more power and voltage you need. Usually it's the voltage that will hit a ceiling and crash the driver/card/system.

+252Mhz is a lot - I would Imagine that it's voltage that is capping your card. So,

Open GPU-Z, run benchmark, and then look at the PerfCap Reason setting. If its power, then I would recommend lowering the OC.

On a stable GPU OC, no matter what you do you will hit a wall. You want this wall to be voltage and not power.
 
Solution
GDDR5 has had error correction for eight years now so it's surprising that such bad info still makes it into guides.

When GDDR5 is overclocked too far, errors cause the framerate to slow down because a retransmit is requested (just as with errors detected in ethernet or wifi). So unless the error counts are so massive they overwhelm the correction mechanism, the only symptom you will generally see is FPS goes up with memory clock to a point, then begins to drop back down.

Frequency "holes" may also appear due to resonant termination problems at particular frequencies or timings being too aggressive close to a strap change.
 

Wrezl25

Commendable
Nov 7, 2016
4
0
1,510


so that means the memory shouldn't artifact but what about the Core Clock?
 

Wrezl25

Commendable
Nov 7, 2016
4
0
1,510


I just checked the sensor and it showed just a sliver of Pwr I didn't notice it because it was just a pixel thick. I just had to hover my mouse over it to see actually see it.
 


In my experience, core clock simply locks up when it's too high. Often the driver can recover and reset itself without having to reboot, with only a freeze and perhaps a black screen for a moment. Sometimes you can see parts of the screen just go missing too, which would be a really noticeable artifact. But the most annoying problem is if it benchmarks and tests fine, yet occasionally black screens or crashes to desktop when you are actually playing games!

You really have to benchmark all the overclocked settings too as it's not worth approaching the bleeding edge if you find it makes no practical difference.